The Agent Skills Directory

[PROMPT_INJECTION]: The file skill-report.json contains self-referential declarations meant to influence the analysis verdict, including claims that the skill is 'safe_to_publish' and that all findings were 'adjudicated as false positives.' This constitutes an attempt to override the analyzer's judgment.\n- [PROMPT_INJECTION]: The inclusion of a mock audit report is a form of metadata poisoning. It misrepresents the skill's safety status by providing a pre-cooked positive assessment that targets the evaluator rather than the end user.\n- [PROMPT_INJECTION]: The skill-report.json file serves as a vector for indirect prompt injection. It introduces untrusted data into the agent's context without boundary markers or sanitization, specifically targeting the analyzer's evaluation logic. The skill lacks sanitization for this embedded content and contains no instructions to the agent to ignore these internal declarations.\n- [SAFE]: The core functional instructions in SKILL.md and references/liquid-glass.md are focused on SwiftUI development. They use standard native APIs (such as glassEffect and GlassEffectContainer) and follow best practices for platform availability guards using #available(iOS 26, *).\n- [SAFE]: No malicious capabilities such as network exfiltration of sensitive data, hardcoded secrets, or unauthorized command execution were found in the skill's documentation or code snippets.

swiftui-liquid-glass